YOLO Algorithm for Ripeness Detection in Greenhouse Cucumbers

Accurate identification and classification of crop maturity are essential for optimizing production efficiency in smart agriculture. This study improves cucumber maturity detection by optimizing the YOLO algorithm, focusing on YOLOv5 and YOLOv8, to address challenges in greenhouse environments. A dataset of over 3,000 labeled images across various growth stages, lighting conditions, and environmental factors was used to train and evaluate the models.The results show that YOLOv8n outperforms the other models, achieving 97% recall and 96% precision in high-confidence intervals, making it the most effective model for complex cucumber maturity detection. YOLOv5n, while slightly lower in performance, excelled in medium to low-confidence ranges, offering a good balance of speed and accuracy. YOLOv8n achieved an overall mAP of 0.469, while YOLOv5n reached 0.424, further highlighting YOLOv8n’s superior performance.This study also explored an ensemble approach combining YOLOv5n, YOLOv5x, and YOLOv8n, which led to improved recall and precision, addressing misclassification and low-confidence detections. These findings highlight the importance of model optimization and data diversity in cucumber maturity detection, suggesting that combined YOLO models provide significant advantages for smart agriculture applications.This research contributes to improving the automation and accuracy of crop maturity classification, offering insights for the future of agricultural automation and more efficient farming practices.